Model Details
This model is a 1B llama3 model pretrained from scratch with torchtitan on fineweb-edu with AdamW optimizer. 20x chinchilla rule for 20B tokens seen.
How to use
import torch
from transformers import pipeline
pipe = pipeline(
"text-generation",
model="kz919/llama3_1b_chinchilla_8132025",
)
print(pipe("The key to life is"))
Downstream Eval
ARC, Hellaswag, Lambda_OpenAI, OpenbookQA, PIQA
lm_eval --model hf --model_args pretrained=kz919/llama3_1b_chinchilla_8142025,dtype="bfloat16",add_bos_token=True --tasks lambada_openai,hellaswag,piqa,arc_easy,arc_challenge,openbookqa --device cuda:7 --batch_size 8
| Tasks | Version | Filter | n-shot | Metric | | Value | | Stderr |
|---|
| arc_challenge | 1 | none | 0 | acc | ↑ | 0.2688 | ± | 0.0130 |
| | none | 0 | acc_norm | ↑ | 0.2875 | ± | 0.0132 |
| arc_easy | 1 | none | 0 | acc | ↑ | 0.6082 | ± | 0.0100 |
| | none | 0 | acc_norm | ↑ | 0.5412 | ± | 0.0102 |
| hellaswag | 1 | none | 0 | acc | ↑ | 0.3459 | ± | 0.0047 |
| | none | 0 | acc_norm | ↑ | 0.4169 | ± | 0.0049 |
| lambada_openai | 1 | none | 0 | acc | ↑ | 0.3311 | ± | 0.0066 |
| | none | 0 | perplexity | ↓ | 38.2983 | ± | 1.5427 |
| openbookqa | 1 | none | 0 | acc | ↑ | 0.2340 | ± | 0.0190 |
| | none | 0 | acc_norm | ↑ | 0.3500 | ± | 0.0214 |
| piqa | 1 | none | 0 | acc | ↑ | 0.6795 | ± | 0.0109 |
| | none | 0 | acc_norm | ↑ | 0.6774 | ± | 0.0109 |
MMLU
| Groups | Version | Filter | n-shot | Metric | | Value | | Stderr |
|---|
| mmlu | 2 | none | | acc | ↑ | 0.2529 | ± | 0.0037 |
| - humanities | 2 | none | | acc | ↑ | 0.2459 | ± | 0.0063 |
| - other | 2 | none | | acc | ↑ | 0.2424 | ± | 0.0077 |
| - social sciences | 2 | none | | acc | ↑ | 0.2697 | ± | 0.0080 |
| - stem | 2 | none | | acc | ↑ | 0.2572 | ± | 0.0078 |